Applied AI architect @ Microsoft. I build AI systems that move a business number, and I build them with the people who will run them.
Every repo here is a working reference architecture: deployable infrastructure, real code, and a write-up explaining the design decisions. Most pair with a deep dive on stochasticcoder.com.
stochasticcoder.com • LinkedIn
Production patterns that move past prompt and response into tool-driven execution, governed autonomy, and measurable outcomes.
-
ACH Remittance Intelligence
Turns messy remittance emails (PDFs, spreadsheets, scans, EDI printouts) into payment data reconciled to the cent, with gated agents for exceptions and per-run AI cost on a dashboard.- Key Patterns: Deterministic first, generative second, verified always · gated agents with cited RAG · Durable Functions orchestration · per-run cost metering
-
Self-Healing Pipelines with Azure SRE Agent
Takes a production failure spike to an automated code fix and a merged GitHub PR, with a human approving only the final merge.- Key Patterns: Agentic reliability · autonomous remediation loops · GitHub Copilot integration
-
Fabric + Foundry Insight & Presentation Agents
Cuts business-review prep from hours to minutes: a Fabric Data Agent answers questions over live data, and Foundry agents generate the PowerPoint deck through MCP.- Key Patterns: Fabric Data Agent interface · multi-turn reasoning over live data · MCP-driven document generation
-
Microsoft Foundry Agentic Workshop
A hands-on workshop for building agentic systems end to end: vector search, multi-agent orchestration with LangGraph, and evaluation.- Key Patterns: Developer enablement · multi-agent evaluation · LangGraph orchestration
-
Travel AI Agent
A full-stack conversational agent that answers traveler questions and completes bookings, with vector search and transactional state.- Key Patterns: Tool-calling agents · vector search · stateful conversations
-
Always-On Contract Risk Platform
Event-driven agents that continuously monitor FX and interest-rate contract risk, with auditable, repeatable evaluations.- Key Patterns: Event-driven autonomy · MCP tool isolation · auditable decision trails
- Microsoft-Decision-1 in a Document Processing Pipeline · Oct 2026
- Decoupling Data and Logic: Using Fabric Data Agents and Foundry to Automate Presentation Generation · Jun 2026
- Beyond the Alert: Building Self-Healing Pipelines with Azure SRE Agent and GitHub Copilot · Apr 2026
- Designing Sequential Multi-Agent Pipelines with Microsoft Foundry · Mar 2026
- Optimizing AI Agents for Scale: Triage, Throttle Control, and Model Right-Sizing · Feb 2026
➡️ More on stochasticcoder.com
⭐ marks the featured patterns above.
| Project | Services | Tech Stack | Pattern & Use Case |
|---|---|---|---|
| BrandSense | Microsoft Foundry, Azure OpenAI, AI Search, APIM | Python, FastAPI, React, Terraform | Multi-agent brand analysis using retrieval, scoring, and validation workflows |
| Contract Risk Platform ⭐ | Microsoft Foundry, AKS, Azure DocumentDB | Python, Bicep, MCP, OpenTelemetry | Always-on, event-driven contract risk monitoring with auditable workflows |
| Travel AI Agent ⭐ | Azure DocumentDB (Cosmos DB for MongoDB vCore) | Python, LangChain, FastAPI, React | Conversational agent with vector search, state management, and transactional workflows |
| Full Stack RAG System | Azure DocumentDB (Cosmos DB for MongoDB vCore) | Python, LangChain, FastAPI, React | Enterprise retrieval-augmented generation pattern |
| Semantic Kernel RAG System | Microsoft Foundry, AI Search, Cosmos DB, Durable Functions | Semantic Kernel, Python, FastAPI, Bicep | RAG orchestration and a foundation for agent-based systems |
| Project | Services | Tech Stack | Pattern & Use Case |
|---|---|---|---|
| ACH Remittance Intelligence ⭐ | Microsoft Foundry, Content Understanding, Durable Functions, AI Search, Service Bus, Container Apps | Python, FastAPI, React, Bicep | Remittance emails → verified payment data, with gated agents and per-run AI cost |
| Self-Healing Pipelines (SRE Agent) ⭐ | Azure SRE Agent, Container Apps, Cosmos DB, Managed Identity | Python, FastAPI, Terraform, GitHub Actions | Incident detection to merged fix PR with autonomous remediation |
| Fabric + Foundry Insight & Presentation Agents ⭐ | Microsoft Fabric, Microsoft Foundry, Container Apps, Managed Identity | Python, FastAPI, FastMCP, python-pptx | Fabric Data Agent → conversational insight → MCP-generated business review decks |
| Microsoft Foundry Agentic Workshop ⭐ | Microsoft Foundry, AI Search, Container Apps, Azure Functions | Python, LangGraph, Semantic Kernel, MCP | Multi-agent orchestration and evaluation patterns |
| ITSM Multi-Agent System | Microsoft Foundry, AI Search, APIM, Managed Identity | Python, FastAPI, React, Terraform | Agent-based ticket classification, routing, and remediation |
| Agents Audit System | Foundry Agent Service, Azure OpenAI, AI Search, Cosmos DB | Python, FastMCP, Bicep | Agentic accounts-payable auditing with RAG, explainable reasoning, and policy compliance |
| Project | Services | Tech Stack | Pattern & Use Case |
|---|---|---|---|
| MCP YARP Gateway | AKS, Cosmos DB, Key Vault, Managed Identity | .NET, YARP, Bicep, Python | Secure MCP reverse proxy with controlled agent tool access |
| Large Document Summarization | Azure OpenAI, Durable Functions, Blob Storage | Python, Bicep, PowerShell | Distributed fan-out/fan-in processing for large documents |
| Microsoft Foundry Deployment | Microsoft Foundry, AI Search, Private Endpoints, Managed Identity | Bicep, PowerShell | Enterprise deployment, networking, and security foundations |
- Outcome first: Every architecture starts with the business case and the number it should move.
- System design over isolated prompts: Prompts are brittle; architectures must be resilient.
- Governed autonomy: Agents operate within explicit guardrails, with humans approving what matters.
- Observable by default: Tracing and evaluation are non-negotiable for production agents.